Cloud cover provides crucial information for many applications such as planning land observation missions from space. It remains, however, a challenging variable to forecast, and numerical weather prediction (NWP) models suffer from significant biases, hence, justifying the use of statistical postprocessing techniques. In this study, ARPEGE (Météo-France global NWP) cloud cover is postprocessed using a convolutional neural network (CNN). CNN is the most popular machine learning tool to deal with images. In our case, CNN allows the integration of spatial information contained in NWP outputs. We use a gridded cloud cover product derived from satellite observations over Europe as ground truth, and predictors are spatial fields of various variables produced by ARPEGE at the corresponding lead time. We show that a simple U-Net architecture (a particular type of CNN) produces significant improvements over Europe. Moreover, the U-Net outclasses more traditional machine learning methods used operationally such as a random forest and a logistic quantile regression. When using a large number of predictors, a first step toward interpretation is to produce a ranking of predictors by importance. Traditional methods of ranking (permutation importance, sequential selection, etc.) need important computational resources. We introduced a weighting predictor layer prior to the traditional U-Net architecture in order to produce such a ranking. The small number of additional weights to train (the same as the number of predictors) does not impact the computational time, representing a huge advantage compared to traditional methods.

Near-surface winds over complex terrain generally feature a large variability at the local scale. Forecasting these winds requires high-resolution numerical weather prediction (NWP) models, which drastically increase the duration of simulations and hinder them in running on a routine basis. Nevertheless, downscaling methods can help in forecasting such wind flows at limited numerical cost. In this study, we present a statistical downscaling of WRF (Weather Research and Forecasting) wind forecasts over southeastern France (including the southwestern part of the Alps) from its original 9 km resolution onto a 1 km resolution grid (1 km NWP model outputs are used to fit our statistical models). Downscaling is performed using convolutional neural networks (CNNs), which are the most powerful machine learning tool for processing images or any kind of gridded data, as demonstrated by recent studies dealing with wind forecast downscaling. The previous studies mostly focused on testing new model architectures. In this study, we aimed to extend these works by exploring different output variables and their associated loss function. We found that there is no one approach that outperforms the others in terms of both the direction and the speed at the same time. Finally, the best overall performance is obtained by combining two CNNs, one dedicated to the direction forecast based on the calculation of the normalized wind components using a customized mean squared error (MSE) loss function and the other dedicated to the speed forecast based on the calculation of the wind components and using another customized MSE loss function. Local-scale, topography-related wind features, which were poorly forecast at 9 km, are now well reproduced, both for speed (e.g., acceleration on the ridge, leeward deceleration, sheltering in valleys) and direction (deflection, valley channeling). There is a general improvement in the forecast, especially during the nighttime stable stratification period, which is the most difficult period to forecast. The result is that, after downscaling, the wind speed bias is reduced from −0.55 to −0.01 m s⁻¹, the wind speed MAE is reduced from 1.02 to 0.69 m s⁻¹ (32 % reduction) and the wind direction MAE is reduced from 25.9 to 15.5^∘ (40 % reduction) in comparison with the 9 km resolution forecast.

In regions of complex topography, local flows are difficult to forecast on a routine basis, especially in stable conditions, due to the coarse resolution of operational models. The Cadarache valley (southeastern France) features this sort of complex topography. The Weather Research and Forecasting (WRF) model is run daily to forecast the weather in this region with a horizontal resolution of 3 km. Such a resolution cannot resolve all topography details of the small Cadarache valley, and therefore its local wind patterns. Other variables, however, that are less dependent on the subgrid topography, are satisfactorily forecasted, and used as inputs to an artificial neural network (ANN) designed to reproduce wind observations inside the valley from WRF forecasts. A variable selection procedure identified 5 key input variables that best drive the ANN. With respect to the WRF output, the ANN significantly improves forecasted low-level winds, both for speed and direction. This study demonstrates the potential for the ANN technique to be used as a correcting tool to forecast weather conditions at the local scale when numerical modeling is performed at a resolution too coarse to take into account the effect of local topography.

We hereby present a new method with which to nowcast a thermally driven, downvalley wind using an artificial neural network (ANN) based on remote observations. The method allows the retrieval of wind speed and direction. The ANN was trained and evaluated using a 3-month winter-period dataset of routine weather observations made in and above the valley. The targeted valley winds feature two main directions (91% of the total dataset) that are aligned with the valley axis. They result from downward momentum transport, channeling mechanisms, and thermally driven flows. A selection procedure of the most pertinent ANN input variables, among the routine observations, highlighted three key variables: a potential temperature difference between the top and the bottom of the valley and the two wind components above the valley. These variables are directly related to the mechanisms that generate the valley winds. The performance of the ANN method improves on an earlier-proposed nowcasting method, based solely on a vertical temperature difference, as well as a multilinear regression model. The assessment of the wind speed and direction indicates good performance (i.e., wind speed bias of −0.28 m s⁻¹ and 84% of calculated directions stray from observations by less than 45°). Major sources of error are due to the misrepresentation of cross-valley winds and very light winds. The validated method was then successfully applied to a 1-yr period with a similar performance. Potentially, this method could be used to downscale valley wind characteristics for unresolved valleys in mesoscale simulations.

In this paper, two different computationally inexpensive methods for nowcasting/data filling spatially varying meteorological variables (wind velocity components, specific humidity, and virtual potential temperature) covering scales ranging from 100 m to 5 km in regions marked by complex terrain are compared. Multivariable linear regression and artificial neural networks are used to predict micrometeorological variables at eight locations using the measurements from three nearby weather stations. The models are trained using data gathered from a system of eleven low-cost automated weather stations that were deployed in the Cadarache Valley of southeastern France from December 2016 to June 2017. The models are tested on two held-out periods of measurements of thermally-driven flow and synoptically forced flow. It is found that the models have statistically significant performance differences for the wind components during the synoptically driven flow period (p = 6.6 × 10⁻³ and p = 2.0 × 10⁻² for U and V, respectively), but perform the same otherwise. These methods can be used to spatially fill gaps in micrometeorological datasets. Recommended future work should include statistically interpreting the predictive models and testing their capabilities on meteorological datasets from different locations.

Ensemble forecast approaches have become state-of-the-art for the quantification of weather forecast uncertainty. However, ensemble forecasts from numerical weather prediction models (NWPs) still tend to be biased and underdispersed, hence justifying the use of statistical post-processing techniques to improve forecast skill. In this study, ensemble forecasts are post-processed using a convolutional neural network (CNN). CNNs are the most popular machine learning tool to deal with images. In our case, CNNs allow to integrate information from spatial patterns contained in NWP outputs.We focus on solar radiation forecasts for 48 hours ahead over Europe from the 35-members ARPEGE (Météo-France global NWP) and a 512-members WRF (Weather Research and Forecasting) ensembles. We used a U-Net (a special kind of CNN) designed to produce a probabilistic forecast (quantiles) using as ground truth the CAMS (Copernicus Atmosphere Monitoring System) radiation service dataset with a spatial resolution of 0.2°.